RAD51 recruitment but not replication fork stability associates with PARP inhibitor response in ovarian cancer patient-derived xenograft models Authors: Francien Talens, Vivian Oliviera Nunes Teixeira, Yannick P Kok, Mengting Chen, Efraim H Rosenberg, Rashmie Debipersad, Evelien W Duiker, Nathalie van den Tempel, Marketa Janatova, Petra Zemankova, Petra M Nederlof, G Bea A Wisman, Zdenek Kleibl, Steven de Jong, Marcel A T M van Vugt Published: 2024 ABSTRACT AbstractPoly(ADP‐ribose) polymerase (PARP) inhibitors (PARPis) are currently used to treat BRCA1/2 mutant cancers. Although PARPi sensitivity has been attributed to homologous recombination (HR) defects, other roles of HR factors have also been linked to response to PARPi, including replication fork protection. In this study, we investigated PARPi sensitivity in ovarian cancer patient-derived xenograft (PDX) models in relation to HR proficiency and replication fork protection. Analysis of BRCA1/2 status showed that in our cohort of 31 ovarian cancer PDX models 22.6% harbored a BRCA1/2 alteration (7/31), and 48.3% (15/31) were genomically unstable as measured by copy number alteration analysis. In vivo, PARPi olaparib response was measured in 15 selected PDX models. Functional assessment of HR using ex vivo irradiation-induced RAD51 foci formation identified all olaparib-sensitive PDX models, including four models without BRCA1/2 alterations. In contrast, replication fork protection or replication speed in ex vivo tumor tissue did not correlate with olaparib response. Targeted panel sequencing in olaparib-sensitive models lacking BRCA1/2 alterations revealed a MUS81 variant as a possible mechanism underlying PARPi sensitivity. Combined, we show that ex vivo RAD51 analysis effectively predicts in vivo olaparib response and revealed a subset of PARPi-sensitive, HR-deficient ovarian cancer PDX models, lacking a BRCA1/2 alteration. Introduction Ovarian cancer is the eighth most common cancer type among women worldwide and survival rates have hardly improved over the last decades (1). High-grade serous ovarian cancer (HGSOC) is the most common subtype of ovarian cancer (∼70%) and is frequently diagnosed in an advanced stage, with the majority of patients being diagnosed with stage >IIB disease. Importantly, patients with metastatic HGSOC have a 5-year relative survival rate of ∼30%. Current treatment of HGSOC combines surgery with (neo)adjuvant platinum-based chemotherapy. Approximately 15–20% of HGSOC are characterized by BRCA1/2 mutations or BRCA1 promoter hypermethylation (2). In line with this notion, females carrying a germline BRCA1 or BRCA2 mutation have a highly increased lifetime risk to develop breast and/or ovarian cancer (3,4). BRCA1 and BRCA2 function in homologous recombination (HR), an error-free pathway to repair toxic DNA double-stranded breaks (DSBs) (5). HR proteins, such as BRCA1 and BRCA2, have additional functions in preventing chromosomal instability, including preventing the degradation of nascent DNA at stalled replication forks (6,7) and suppressing the formation of single-stranded (ss)DNA gaps in the lagging strand during DNA replication (8–10). Tumor cells with mutations in these HR factors are sensitive to treatment with inhibitors of the Poly(ADP‐ribose) polymerase (PARP) enzyme, which functions in single-strand break (SSB) repair (11,12). Several PARP inhibitors (PARPis), including olaparib, niraparib and rucaparib, have been clinically evaluated and treatment with these drugs improved progression-free survival in patients with BRCA1/2 mutated ovarian cancer and therefore resulted in FDA approval. Currently, the PARPi olaparib is approved for use in BRA1/2 mutant ovarian and BRCA1/2 mutant HER2-negative metastatic breast cancers (13), whereas niraparib is approved for all ovarian cancers that responded to platinum-based therapy. Additionally, olaparib and rucaparib are indicated for castration-resistant BRCA1/2 mutant prostate cancer. How exactly PARPis exert their cytotoxic effects is a topic of ongoing debate. Initial hypotheses involved the accumulation of ssDNA breaks, which lead to DSBs at sites of ongoing replication (11,12,14). As a logical consequence in this model, the formation of DSBs would necessitate BRCA1/2 and related factors for accurate repair through HR. Subsequently, PARPis were found to trap PARP molecules onto DNA and perturb DNA replication (15), and that BRCA1/2 prevent the nucleolytic degradation of stalled replication forks (7). Combined, these findings gave rise to a model in which PARP inhibition leads to stalled replication forks that require functional BRCA1/2 for protection and that the sensitivity to PARPis is determined by the ability of cells to protect stalled replication forks (16,17). A more recent study revealed that PARP1 slows down replication speed (18), and that accelerated replication velocity induced by PARP inhibition leads to DNA lesions. Along this line, BRCA1/2 was shown to prevent PARPi-induced ssDNA gaps during replication (19). Moreover, levels of ssDNA gaps, rather than DSB accumulation, were shown to associate with PARPi response (20). Finally, PARP1 was demonstrated to protect the replisome from transcription-replication conflicts, and conversely, PARP inhibition induces replication-transcription conflicts that underlie cytotoxicity in HR-deficient cancer cells (21). Also several mechanisms have been described by which HR-defective tumors become resistant to PARPi treatment. Reversion mutations in BRCA1/2 form a main mechanism of acquired resistance in the clinic (22). Also, mutations in other genes have been shown to confer resistance to PARP inhibition. For instance, inactivation of TP53BP1 (23,24), REV7 (25) or Shieldin complex genes (26,27) can restore HR in BRCA1 mutant cancer cells and leads to PARPi resistance in experimental models. Similarly, mutations in the CST genes rescue HR in BRCA1 mutant cancer cells and lead to PARPi insensitivity (28). Because inactivation of the above-mentioned genes partially restores HR, these findings pointed at defective HR as a determinant of PARPi sensitivity. However, loss of 53BP1 also rescues Okazaki fragment processing and ssDNA gap formation, providing an alternative explanation for the observed PARPi resistance (19). In addition, mutations in PAXIP2 were described to rescue replication fork stability in BRCA2 mutant cells and thereby determine PARPi sensitivity (16). These latter observations suggest that mechanisms beyond HR determine PARPi sensitivity. Of note, a separation-of-function mutant of Brca2 that is HR proficient but cannot suppress ssDNA gaps or mediate fork protection does not confer PARPi sensitivity in vivo, while a Brca2 mutant that is defective in HR does lead to PARPi sensitivity (29), again pointing at HR as the main determinant of PARPi sensitivity. A solid understanding of the mechanisms that drive PARPi sensitivity and resistance is needed to effectively identify PARPi eligible patients. Currently used homologous recombination deficiency (HRD) tests to select patients for PARPi treatment are based on BRCA1/2 status or genomic scars (30). For example, the myChoice HRD test combines BRCA1/2 mutation status with different measurements of genomic instability in tumor cells, including loss-of-heterozygosity (LOH), telomeric imbalance and large-scale state transitions, and has demonstrated varying results in predicting platinum response in triple-negative breast cancer (TNBC) (31,32). Intriguingly, this test did not predict response to the PARPi niraparib in platinum-sensitive ovarian cancer (33). In a different approach, algorithms were developed on whole-genome sequencing profiles to distinguish BRCA1/2-mutated breast cancers from sporadic breast cancer (34,35). Of these, the HRDetect algorithm was able to detect HR-deficient tumors without BRCA1/2 mutations in different cancer types and HRDetect scores were associated with response to platinum-based chemotherapy in advanced breast cancer, independently of BRCA1/2 mutation status (36). These approaches are highly relevant, because PARPi treatment is expected to also be effective in tumors that lack a BRCA1/2 mutation but with mutation or epigenetic silencing of other HR genes, including PALB2, CHEK1/2, RAD51C, RAD51D and ATM (2,37–40). In line with this notion, several clinical trials have shown proven efficacy of PARPi beyond BRCA1/2-mutated ovarian cancer (33,41). Based on these findings, all patients with HRD tumors, also those caused by mutations other than BRCA1/2 alterations, are currently eligible for PARPi treatment. Yet, proper selection methods are needed to select those patients with high likelihood for treatment benefit. Since BRCA1/2 mutation analysis nor analysis of genomic features associated with HRD per se reflects actual HR functionality, these assays may not be effective in proper selection of patients that benefit from PARPi treatment. Therefore, an ex vivo functional test for HR pathway proficiency was developed, the RECAP assay, which assesses foci formation of the RAD51 recombinase (42). The loading of RAD51 onto ssDNA at DSBs is a crucial step in HR and can be visualized with immunofluorescence (43). Although this approach does not take into account all functions of HR proteins, including replication fork protection or ssDNA suppression, this assay successfully identified HRD breast cancers that lacked a BRCA1/2 mutation but were not able to recruit RAD51 foci and were therefore considered HRD (44,45). Furthermore, formation of RAD51 foci correlated to PARPi resistance in BRCA1/2 mutant breast cancer models, regardless of the underlying resistance mechanism (46,47). Also, functional HR testing could predict platinum sensitivity in ovarian cancer patients (48). In primary cultures derived from ascitic fluid from ovarian cancer patients, RAD51 foci formation correlated to in vitro PARPi response (49). However, it remains unclear whether in vivo responses to PARPi can be predicted using functional HR assays in ovarian cancer models. Morerover, other functions of BRCA1/2 such as replication fork protection have to date not been included in functional assays to predict PARPi response in clinically relevant cancer models. In this study, we performed a functional assessment of HR using the RAD51-based RECAP assay along with functional analysis of replication fork protection in a cohort of patient-derived xenograft (PDX) models of ovarian cancer. These parameters were correlated to in vivo olaparib sensitivity in PDX models of ovarian cancer. We show that the RECAP assay is superior in identifying olaparib-sensitive PDX models, when compared to BRCA1/2 status, copy number alteration (CNA) profiles and replication fork stability. These results demonstrate that clinical evaluation of functional HR assessment is warranted to select patients with ovarian cancer for PARPi treatment. Materials and methods Patient-derived tumor xenograft (PDX) mouse models Cryopreserved tumor pieces from a cohort of ovarian cancer patients were thawed for implantation into 6- to 10-week-old female NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice (internal breed, Central Animal Facility, University Medical Center Groningen), as described previously (50). Typically, one tumor piece was subcutaneously implanted on one side of the flank and surgery was performed as previously described (50). Mice received sterilized food and water ad libitum and were kept under pathogen-free conditions in the Central Animal Facility at the University Medical Center Groningen. Animal experiments were approved by the Institutional Animal Care and Use Committee of the University of Groningen and followed the EU Guideline on Animal Experiments. Mouse experiments were divided into two phases: an expansion phase and a treatment phase. For the expansion phase, two mice were implanted with F1 or F2 ovarian cancer specimens. When tumors reached a volume of ∼1000 mm3, mice were terminated and tumors were used for implantation into 14 recipient mice for the treatment phase. In parallel to implantation for treatment studies, tumor pieces were used for ex vivo DNA fiber assay analysis and the RECAP assay. Analysis of copy number alterations (CNAs) and genomic status of BRCA1/2 For analysis of CNAs and BRCA1/2 status, frozen tissue slices (10 μm) were cut from F1 tumor material that was stored at −80°C. Tissue slices were stained for hematoxylin and eosin (H&E), and were analyzed by a gynecologic pathologist to determine tumor cell percentage. Genomic DNA was isolated using the QIAamp DNA mini kit (Qiagen) according to manufacturer’s protocol. For BRCA1/2 mutational analysis, samples were sequenced using Multiplicom BRCA MASTR for NGS. To determine BRCA1 promoter methylation, MS-MLPA (SALSA MLPA Probemix ME001, MRC Holland) was performed as described. Cartagenia Bench was used for variant calling and results were compared with the Dutch national BRCA1/2 database, incorporated in the Leiden Open Variation Database LOVD v.3.0 (51). Low-coverage whole-genome sequencing and BRCA1/2-like classification was performed as described previously (52). Samples were classified as BRCA-like if the predicted probability was >0.5 and non–BRCA-like if the predicted probability was ≤ 0.5. BRCA1 and BRCA2 scores are listed in Supplementary Table S1. Based on CNAs profiles, models were classified by visual interpretation into five categories (A–E), ranging from ‘flatliner’ profiles containing no CNAs (category A), few aberrations (categories B and C) and profiles with many aberrations and intrachromosomal rearrangements (categories D and E). Examples of CNA profiles can be found in Supplementary Figure S1. MSI analysis Microsatellite instability (MSI) was assessed by testing tumor DNA in a pentaplex polymerase chain reaction (PCR)-based assay using fluorescent labelled primers of five mononucleotide repeat targets (BAT25, BAT26, NR24, NR21, NR27), followed by fragment analysis. In case of instability in two markers or more, the tumor is defined as MSI-high. Instability of only one marker is considered MSI-low. When five markers are without instability the tumor is defined as MS stable (MSS). In vivo evaluation of PARPi response Mice were closely observed, weighted and tumor size was measured at least once a week. Tumor size was measured using a caliper, and volume was calculated using the following formula: (width2 × length)/2. When tumors reached a volume of ∼200 mm3, mice were distributed into a vehicle treatment group (10% dimethyl sulfoxide [DMSO], 10% 2-hydroxypropyl-β-cyclodextrin (HPBCD) in phosphate saline buffer [PBS]) or olaparib treatment group (100 mg/kg diluted in 10% DMSO, 10% HPBCD in PBS). Treatments were administered through intraperitoneal injection, comparable to other olaparib in vivo PDX studies (53). Mice were divided into treatment groups based on tumor size to maintain a comparable mean tumor size in both groups at start of treatment. Treatment was administered using intraperitoneal injections for six times a week. After 28 days, mice were terminated by cervical dislocation under isoflurane anesthesia. Tumors were harvested, weighted and cut into two pieces: one piece was snap frozen and kept in −80°C whereas the other piece was stored in formalin. Following the guidelines of animal experimentation, the following humane endpoints were followed to avoid unnecessary suffering of animals: tumor size > 1500 cm3, weight loss > 15%, ulceration of tumors, observation of hunched posture or altered behavior. Tumor growth (%) during treatment was calculated by comparing the tumor volume at the end of treatment to tumor volume at day 0. The response to olaparib for each PDX model was calculated by comparing the mean tumor growth in the olaparib-treated group with the mean tumor growth in the DMSO-treated group as a percentage. Adapted Response Evaluation Criteria in Solid Tumors (RECIST) guidelines were used to classify PDX models as responders to olaparib (54). A model was considered sensitive when a decrease of at least 30% (partial response, RECIST) in mean tumor growth upon olaparib treatment was observed compared to DMSO treatment. Functional RAD51 (RECAP) assay The RECAP (REpair CAPacity) assay, a functional assay to assess RAD51 foci formation in tumor tissue after ex vivo irradiation, was performed as described previously (42). Fresh PDX tumor tissue harvested from mice was cut into 2–3 mm pieces and placed in 6-well plates with Roswell Park Memorial Institute (RPMI) medium, supplemented with 10% fetal calf serum (FCS) and 1% penicillin/streptomycin (P/S) within 4 h after resection. After irradiation using a Cesium137 source (5 Gy, IBL 637 Cesium137 gamma-ray machine), tumor pieces were incubated for 3 h at 37°C. Tumor pieces were then put in formalin overnight and imbedded in paraffin. Tissue sections were deparaffinized with xylene and hydrated with decreasing concentrations of ethanol. For antigen retrieval, sections were microwaved for 12 min at 100°C in retrieval solution (DAKO, #S2367). Sections were cooled down for 20 min, permeabilized with 0.2% Triton-X-100 in PBS for 20 min at room temperature, washed once with PBS and incubated with DNAse I (1000 U/ml, Roche, #04 536 282 001) for 1 h at 37°C in an incubator. Slides were incubated in blocking buffer (1% BSA, 2% FCS in PBS) for at least 30 min. Primary antibodies were diluted in blocking buffer and incubated at room temperature for 1 h. Primary antibodies used were mouse anti-human RAD51 (1:200; Genetex, #gtx70230 clone 14B4) and rabbit anti-human Geminin (1:400, Protein Tech, #10 802). Secondary antibodies AlexaFluor 594-conjugated goat anti-mouse (1:1000, Invitrogen, #A11005) and AlexaFluor 488-conjugated goat anti-rabbit (1:1000, Invitrogen, #A11034) were diluted in blocking buffer and incubated for 1 h at room temperature. Sections were air dried in the dark for 30 min and mounted with ProLong Diamond Antifade Mountant reagent with DAPI (4′,6-diamidino-2-phenylindole, Invitrogen, #P36966) and stored overnight at 4°C. RAD51 staining was quantified by scoring the percentage of geminin-positive cells with ≥5 foci/cell. Tumors with <20% of geminin-positive cells with ≥5 RAD51 foci/cell were classified as HRD, as has been used in other studies (42,44,55). At least 30 geminin-positive cells were analyzed on at least two different slides. Immunofluorescence images were acquired on a Leica DM-6000RXA microscope using LAS X software. DNA fiber assay Fresh tumor material harvested from mice was cut into pieces of ∼2–3 mm each, while kept in RPMI medium supplemental with 10% FCS and 1% P/S. Tumors were dissociated into single-cell suspensions using the Human Tumor Dissociation Kit, according to manufacturer’s instructions (Macs Miltenyi Biotec, #130–095-929). Human cells were enriched from the cell pool using the Mouse Cell Depletion Kit (Macs Miltenyi Biotec, # 130–104-694). After isolation, tumor cells were incubated in 6-well plates and resuspended in RPMI medium supplemented with 10% FCS and 1% P/S, containing CldU (25 μM) for 30 min at 37°C. After extensive washing with medium, cells were incubated in medium supplemented with IdU (250 μM) for 45 min at 37°C. If indicated, IdU was then washed out, and left untreated or treated with hydroxyurea (HU, 5 mM) for 3 h at 37°C. Cells were subsequently lysed on glass slides in lysis buffer (200 mM Tris–HCl, pH 7.4, 50 mM EDTA, 0.5% SDS) and DNA fibers were spread by tilting the glass slides ∼15 degrees. At least three slides of each condition were made. Fibers were fixed with methanol/acetic acid (3:1) for 10 min and then stored at 4°C. Prior to immunolabeling, slides were washed with water and incubated with 2.5 M HCl for 75 min to denature the DNA. Slides were then washed with PBS and incubated with blocking solution (1% bovine serum albumin [BSA], 0.1% Tween-20 in PBS) for 30 min. Next, slides were incubated with primary antibodies rat-anti-BrdU (1:1000, Abcam #ab6326) and mouse-anti-BrdU (1:250, BD Biosciences #347 580) for 1 h, followed by incubation with secondary antibodies AlexaFluor 488-conjugated goat anti-rat (1:500, Invitrogen #A11006) and AlexaFluor 594-conjugated goat anti-mouse (1:500, Invitrogen #A11005) for 90 min. Immunofluorescence images were acquired on a Leica CTR6000 microscope using LAS X software (Leica Application Suite X). Fiber length was measured using ImageJ software. A conversion factor of 1 μm = 2 kb was used, as described previously (56). EdU/cytokeratin immunofluorescence staining Single-dissociated cells used for DNA fiber analyses were incubated with EdU for 1 h at 37°C. Cells were harvested and cytospinned on glass slides. Cells were fixed with 3.7% formaldehyde in PBS for 15 min at room temperature. Until staining, slides were stored in −80°C. For EdU staining, cells were permeabilized with 0.5% Triton-X-100 in PBS for 20 min. Subsequently, cells were incubated with a reaction cocktail for 30 min in the dark consisting of 116.3 mM Tris–HCl pH 8.5, 100 mM CuSO4, Alexa fluor azide (0.24 μl per reaction) and 100 mM ascorbic acid in H2O. Washing steps were performed with 3% BSA in PBS. For further staining, cells were washed with PBS and permeabilized with 0.1% Triton-X-100 in PBS for 5 min. Blocking was performed in 2.5% BSA–0.05% Tween-20 in PBS for 1 h. Slides were incubated with primary antibody anti-pan cytokeratin (1:50, AE1AE3, DAKO) in 0.05% Tween-20 in PBS overnight at 4°C followed by secondary Alexa-conjugated antibodies for 1 h at room temperature. Immunofluorescence images were acquired on a Leica CTR6000 microscope using LAS X software (Leica Application Suite X). Targeted DNA sequencing Multi-gene panel sequencing of DNA from F1 PDX tumor material was performed using the CZECANCA panel (57). The procedure was performed as described previously with the following minor modifications. The CZECANCA panel version 1.2 was used, which targets 226 genes instead of 219 genes. The list of targeted genes included in CZECANCA panel version 1.2 can be found in Supplementary Table S2. Sequencing was performed on the Illumina NextSeq platform. Coverages exceeded 500× for the majority of targeted sequences. In total, 39 341 variants were called and the following filters were applied: sequencing errors/low quality variants (quality < 150 from GATK software pipeline), frequent variants (mutation allele frequency, MAF > 0.05 in 1000 Genomes project and ESP6500), variants with difficult interpretation (extragenic, intronic [except splicing alterations], UTR), variants described in ClinVar as benign or likely benign, or presented in non-cancer and general population controls. Remaining 22 variants were validated using the Integrative genomics viewer (IGV) and are listed in Table 2. Table 2.Variants found with panel sequencing in PARPi sensitive PDX modelsPDX #sample # (see table 1)ChromosomeGeneAlterationhet/homExogenic function37S01chr10DMBT1DMBT1:NM_007 329:exon17: c.1975C > T:p.Q659XhomStopgain SNV37S01chr11MUS81MUS81:NM_025 128:exon14: c.1487G > A:p.R496QhomNonsynonymous SNV37S01chr5TERTTERT:NM_198 253:exon2: c.1171C > T:p.P391ShetNonsynonymous SNV37S01chr9TLR4TLR4:NM_138 554:exon2:c.T197A:p.L66Q, TLR4:NM_003 266:exon3:c.T77A:p.L26QhetNonsynonymous SNV37S01chr17TP53TP53:NM_000 546:exon7: c.742C > T:p.R248WhomNonsynonymous SNV37S01chr10ZNF365ZNF365:NM_014 951:exon5: c.1093G > A:p.E365KhetNonsynonymous SNV67S02chr7EGFREGFR:NM_201 284:exon16: c.2060G > A:p.S687NhetNonsynonymous SNV67S02chr8PREX2PREX2:NM_024 870:exon17: c.1876G > A:p.E626KhetNonsynonymous SNV67S02chr17TP53TP53:NM_000 546:exon8: c.844C > T:p.R282WhomNonsynonymous SNV81S03chr17BRCA1BRCA1:NM_007 294:exon10: c.1823delA:p.K608fshetFrameshift deletion174S05chr11CWF19L2CWF19L2:NM_152 434:exon3: c.273delA:p.K91fshetFrameshift deletion174S05chr10DMBT1DMBT1:NM_007 329:exon26: c.2975C > T:p.A992VhetNonsynonymous SNV174S05chr9TLR4TLR4:NM_003 266:exon4: c.1402C > A:p.L468MhetNonsynonymous SNV174S05chr17TP53TP53:NM_000 546:exon9: c.972dupT:p.G325fshomFrameshift insertion176S06chr11CDKN1CCDKN1C:NM_000 076:exon1: c.549_554del:p.183_185delhomNonframeshift deletion176S06chr10FAM175BFAM175B:NM_032 182:exon9: c.1120G > A:p.D374NhetNonsynonymous SNV176S06chr17TP53TP53:NM_000 546:exon5: c.488A > G:p.Y163ChomNonsynonymous SNV177S07chr13BRCA2BRCA2:NM_000 059:exon10: c.1457delA:p.Q486fshomFrameshift deletion177S07chr17TP53TP53:NM_000 546:exon8: c.817C > T:p.R273ChomNonsynonymous SNV189S08chr13BRCA2BRCA2:NM_000 059:exon23: c.9090delA:p.T3030fshetFrameshift deletion193S09chr16AXIN1AXIN1:NM_003 502:exon2: c.833C > T:p.P278LhetNonsynonymous SNV193S09chr13ERCC5ERCC5:NM_000 123:exon15: c.3356C > T:p.A1119VhetNonsynonymous SNV193S09chr17TP53TP53:NM_000 546:exon7: c.737T > C:p.M246ThomNonsynonymous SNV195S10chr17BRCA1BRCA1:NM_007 294:exon23: c.5542delC:p.Q1848fshetFrameshift deletion Variants found with panel sequencing in PARPi sensitive PDX models MUS81 knockout generation in U2OS The human osteosarcoma cell line U2OS was obtained from ATCC (#HTB-96) and was cultured in Dulbecco’s Minimum Essential Medium (DMEM, Thermofisher), supplemented with 10% FCS, 1% penicillin and 1% streptomycin (Gibco). A single-guide RNA (sgRNA) targeting MUS81 (GTTTAGGGAAGTGACCCAGG) was cloned into lentiCRISPRv2 puro (Addgene #98 290). U2OS cells were transiently transfected with sgMUS81 using Fugene (Promega), according to manufacturer’s protocol. Two days later, cells were selected by puromycin (1 μg/ml, Sigma) for 4 days and plated in 96-well plates to generate single clones. MUS81 variant analysis eHAP1 parental and MUS81KO cells were a kind gift from Dr S. West and K. Fugger (Crick Institute, London, UK) (58), and were cultured in Iscove Modified Dulbecco Media (IMDM, Thermofisher), supplemented with 10% FCS, 1% penicillin and 1% streptomycin (Gibco). pLNCX2-GFP-MUS81-wt was synthesized by Eurofins Genomics (Germany). eHAP1 cells were transduced with pLNCX2-GFP-MUS81-wt or pLNCX2-GFP-MUS81-R496Q. The R496Q mutation was introduced in pLNCX2-GFP-MUS81-wt by mutagenesis PCR, using the following primers: forward: ′TGGGGAGAAGGCAGCAGCCCTGGTGGATCAATACAGCACCC′ reverse ′GGGTGCTGTATTGATCCACCAGGGCTGCTGCCTTCTCCCCA′. HEK293T cells were transfected with pLNCX2-GFP-MUS81-wt or R496Q along with lentiviral packing plasmids as described previously. Virus was harvested at 48 h after transfection for transduction. eHAP1-MUS81KO cells were transduced for 24 h. At 48 h after transduction of eHAP1-MUS81KO, GFP-positive cells were sorted using a Sony SH800 flow cytometry cell sorter. MTT assays U2OS WT and MUS81 knockout cells were plated in 96-well at 500 cells per well and allowed to attach for 24 h. Olaparib or cisplatin was added to U2OS cells at indicated concentrations for 3 days. eHAP1 cells were plated in 96-well at 500 cells per well and allowed to attach for 24 h. Olaparib was added to eHAP1 cells at indicated concentrations for 4 days. Next, cells were incubated with methylthiazol tetrazolium (MTT, final concentration 0.5 mg/ml) for 3 h. After removal of medium, formazan crystals were dissolved in DMSO. Absorbance was measured at 520 nm, and was quantified with a Benchmark III spectrophotometer (Bio-Rad). Results represent n = 3 independent replicates for U2OS cells and n = 5 independent replicates for eHAP1 cells. Background-ground corrected MTT conversion data were normalized to DMSO treated conditions of wt-eHAP1 cells or wt-U2OS cells. For analysis of area under the curve (AUC), default setting of AUC analysis in Graphpad Prism (version 8.4.2) were used on non-fitted curves and equally spaced X-axis values. RAD51 staining U2OS WT and MUS81 knockout cells were seeded on glass coverslips. On the next day, cells were irradiated (5 Gy) using a CIS international/IBL 637 cesium137 source. At 3 h after irradiation, cells were fixed with 2% paraformaldehyde, permeabilized in 0.1% Triton X-100 in PBS, blocked in 4% BSA in PBS and incubated with primary antibodies against RAD51 (GeneTex, #gtx70230, 1:200) and Geminin (Proteintech, #10 802, 1:200) overnight. Secondary antibodies Alexa-488 (anti-mouse, 1:500) and Alexa-647 (anti-rabit, 1:500) were used and slides were stained with DAPI. Images were made using a Leica DM6000B microscope with a ∼63x immersion objective. Statistical analysis All statistical tests were performed using GraphPad Prism version 7.0. Parametric data were analyzed using unpaired t-tests (two-tailed), one-way ANOVA or two-way ANOVA. Non-parametric data were analyzed using Mann–Whitney tests. Data are presented as means with standard error of the mean (SEM) or median with percentiles for non-parametric data. Correlations were calculated using Spearman’s test. Results Characterization and selection of ovarian cancer PDX models To study the response of ovarian cancers towards the PARPi olaparib, we studied 31 pre-established ovarian PDX models (50). Based on retrieved pathology reports of the primary tumor, 21 out of 31 PDX models represented serous ovarian cancer, of which 20 were HGSOC. Approximately 10 out of 31 PDX models represented other ovarian cancer subtypes, including endometrioid adenocarcinoma (n = 6), carcinosarcomas (n = 2) or mucinous adenocarcinoma (n = 1) (Table 1). Out of 31 PDX models, 7 models harbored a pathogenic BRCA1 or BRCA2 alteration (Table 1). Of these models, PDX177 and PDX203 had a BRCA2 deletion with a high allele frequency, well above 50% (Table 1). The other BRCA1/2 alterations had allele frequency of ∼50%, and were considered heterozygous, without LOH. In addition to the observed genomic alterations, PDX84 showed BRCA1 promoter hypermethylation, which is a commonly observed somatic mechanism of BRCA1 gene silencing. Table 1.Tumor characteristics of PDX models included in the study. Highlighted models were included for in vivo PARPi sensitivityPDX #Tumor typeGradeStageTissue originFollow-up BRCA1/2 status CNV profileMSI statuspanel sequencing (sample #)BRCA1/2 score (see also Supplementary Table S1)Olaparib responseRAD51 assay (RECAP)30Serous cystadenocarcinomaHighIIICDiagnostic laparotomyRelapse after 24 months, deceasedDMSSB2-like37Serous cystadenocarcinomaHighIIICDiagnostic laparotomyNo signs of recurrence (2015)EMSSS01B2-likesensitiveDefective56Serous cystadenocarcinomaHighIIICInterval debulkingNo signs of recurrence (2015)DMSSB1-likeInsensitiveProficient60Serous cystadenocarcinomaHighIIADiagnostic laparotomyDeceased, 17 monthsCMSSB2-like61Mucinous adenocarcinomaUndifferentiatedIVDiagnostic laparotomyDeceasedCMSS67Serous cystadenocarcinomaHighIIICDiagnostic laparotomyNo signs of recurrence (2015)EMSSS02B1 and B2-likeSensitiveDefective68Ovarian carcinosarcomaHighIICDiagnostic laparotomyRecurrence 29 months after last chemo. Re-debulkin. Deceased 40 monthsAMSSInsensitiveProficient70Serous adenocarcinoma with partial oxyfile clearcell componentHighIIIADiagnostic laparotomyDeceasedEMSSB2-like79Serous cystadenocarcinomaLowIIICDiagnostic laparotomyBad response: Palliative situationBMSS81Endometrioid adenocarcinomaModerateICDiagnostic laparotomyNo signs of recurrenceBRCA1: Deletion c.1823delA: p.K608fs in exon 11 of 50%AMSI-HS03InsensitiveProficient84Serous cystadenocarcinomaHighIVDiagnostic laparotomyNo signs of recurrence (2015)BRCA1: promotor methylationEMSSB1 and B2-likeSensitiveDefective102UnknownDiagnostic laparotomyUnknownAMSS112Endometrioid adenocarcinomaHighIIICDiagnostic laparotomyNo signs of recurrenceAMSSS04InsensitiveProficient130Serous cystadenocarcinomaHighIIIBDiagnostic laparotomyNo signs of recurrenceCMSS143Serous cystadenocarcinomaHighIIICInterval debulkingDeceased 11 monthsEMSSB1-likeInsensitiveProficient157Endometrioid adenocarcinomaHighIVDiagnostic laparotomyDeceased 6 monthsBMSSB1-like167Endometrioid adenocarcinomaModerateIAInterval debulkingRecurrence 50 monthsBMSS171Serous cystadenocarcinomaHighIIIRecurrence 2010Recurrence 38 months, deceased 51 monthsAMSS174Serous cystadenocarcinomaHighIVInterval debulkingGood responseDMSSS05B2-likeSensitiveDefective176Serous cystadenocarcinomaIIICDiagnostic laparotomyNo signs of recurrence. Deceased 9 monthsDMSSS06SensitiveDefective177Serous cystadenocarcinoma with partial clear cell differentiationHighIICDiagnostic laparotomyn.a.BRCA2: Deletion c.1457delA: p.Q486fs in exon 10 of 97%DMSSS07B2-likeSensitiveDefective179Serous cystadenocarcinomaHighIIICInterval debulkingInterval: good clinical response, deceased 9 monthsBMSS187Serous cystadenocarcinomaHighIIICInterval debulkingInterval: good responseDMSSInsensitiveProficient188Serous cystadenocarcinomaHighIIICPrimary debulkingProgression within 6 months after chemotherapy (deceased)CMSSB2-like189Endometrioid adenocarcinomaHighICDiagnostic laparotomyDisease free at last follow-up (10 months)BRCA2: Deletion c.9097delA: p.T3033fs in exon 23 of 51%BMSI-HS08191Ovarian carcinosarcomaIIICPrimary debulkingRecurrence 20 monthsDMSSB2-likeInsensitiveProficient193Serous cystadenocarcinomaHighIIICDiagnostic laparotomyAdjuvant chemotherapyBRCA1: Exon 3 duplicationEMSSS09B1 and B2-likeSensitiveDefective195Serous cystadenocarcinomaHighIVDiagnostic laparotomyComplete debulking, adjuvant chemotherapyBRCA1: Deletion c.5542del: p.Q1848fs in exon 24 of 51%DMSSS10B1 en B2-likeSensitiveDefective203Serous cystadenocarcinomaHighIIICPrimary debulkingDisease-free at last follow-up (25 months)BRCA2: Deletion c.7007_7007 + 1delinsTT, a splice mutation in exon 13 of 75%DMSSB1 en B2-like207Serous cystadenocarcinomaHighIIICDiagnostic laparotomyDisease-free at last follow-up (36 months)AMSS208Endometrioid adenocarcinomaLowIIBDiagnostic laparotomyDisease-free at last follow-up (18 months)AMSSAbbreviations: MSI, microsatellite instable; MSS, microsatellite stable.Color scheme:Gray boxes: selected for in vivo analysis for olaparib sensitivity Tumor characteristics of PDX models included in the study. Highlighted models were included for in vivo PARPi sensitivity Abbreviations: MSI, microsatellite instable; MSS, microsatellite stable. Color scheme: Gray boxes: selected for in vivo analysis for olaparib sensitivity As a readout for genomic instability, CNA levels were determined through CNVseq (Supplementary Figure S1A). PDX models were classified in five categories ranging from ‘flatliner’ profiles and profiles that contained very few CNAs (profile A) toward genomically unstable profiles with high amounts of CNAs (profile E). Classification was done without knowledge about other tumor characteristics, such as gene alterations and tumor type. In total, 15 models were considered genomically unstable, based on a CNA profile D or E. Notable, not all PDX models with BRCA1/2 alterations showed high levels of CNAs (i.e. PDX81, PDX189), in line with these models not representing HGSOC but endometrioid tumors, and not having homozygous BRCA1/2 alterations, likely caused by high microsatellite instability (MSI-high) (Table 1). As BRCAness correlates with high levels of genomic instability and is associated with PARPi response, all tumor models with D or E CNA profiles were selected for in vivo assessment of olaparib sensitivity (34). Additionally, two ovarian cancer PDX models without BRCA1/2 alterations and genomically stable profile (PDX68 and PDX112) were included for reference (Supplementary Figure S1B). Furthermore, all models that harbored a BRCA1/2 alteration were included, regardless of their CNAs profile. Approximately 4 out of 19 PDX models that were selected for in vivo PARPi response assessment could not be analyzed, either due to limited tissue availability (PDX203), no tumor development after re-implantation (PDX70, PDX189) or absence of human tumor cells in PDX (PDX30). Ultimately, 15 PDX models were included for in vivo determination of PARPi sensitivity and further ex vivo analyses (highlighted in Table 1). PARPi response in relation to BRCA1/2 status or levels of genomic instability The selected 15 ovarian cancer PDX models were treated for 28 days with olaparib or a solvent control treatment. In total, 8 out of 15 models showed a response to olaparib, as defined as a 30% lower mean tumor volume at the end of treatment in the olaparib-treated group versus the mean tumor growth in the control-treated group (Figure 1A and Table 1). As expected, both PDX models with CNA profile A and lacking BRCA1/2 alterations (PDX68 and PDX112), did not respond to olaparib (Figure 1A and B). For models categorized as genomically unstable (CNA profiles D and E), 8 responded to olaparib, whereas 4 did not (Figure 1A and B). Of the five PDX models with BRCA1/2 alterations, 1 model (PDX81) did not respond to olaparib, likely explained by its endometrioid subtype, lack of BRCA1 LOH, and a corresponding low level of genomic instability (CNA profile A), in line with this model being MSI-high (MSI-H, Table 1). Tumor growth curves and final tumor growth percentages upon olaparib or control treatment for individual PDX models can be found in Supplementary Figure S2 and S3. Response to olaparib in BRCA1/2 mutant/defective or genomically instable ovarian PDX models. (A) Mice were treated for 28 days with olaparib or solvent (DMSO). Bars show the mean tumor size at day 28 in the olaparib-treated group relative to the mean tumor size at day 28 in the solvent-treated group. Each dot represents one olaparib-treated mouse (n ≥ 4). Data are shown as mean ± SEM of olaparib-treated mice. Boxes show characteristics of each PDX model. (B) Representative tumor growth curves of individual PDX models treated with olaparib (blue) or solvent (gray). Percentages of tumor growth were calculated as tumor volume at day x compared to tumor volume at day 0. Data are shown as mean ± SEM of at least 4 mice per group. P-values were calculated using two-way ANOVA with Bonferroni correction; *P < 0.05, **P < 0.01 and ***P < 0.001. (C) Mean tumor size at day 28 in olaparib-treated mice was normalized to mean tumor size in solvent-treated mice as a measure of olaparib response (<100%). Every dot represents mean response one PDX model. BRCA1/2 wt PDX models were separated from PDX models with a BRCA1/2 alteration. Data are shown as mean ± SD of individual PDX models. P-value was calculated using two-tailed Student’s t-test. (D) Olaparib response was calculated as described for panel (C). Each dot represents one PDX model. PDX models with CNA profile A were separated from PDX models with CNA profile D or E. Data are shown as mean ± SD of different PDX models. P-value was calculated using two-tailed Student’s t-test. (E) Olaparib response was calculated as described for panel (C). Each dot represents one PDX model. PDX models classified as ‘BRCA1-like’ were separated from non-‘BRCA1-like’ tumors based on BRCA1-classifier outcome. Data are shown as mean ± SD of different PDX models. P-value was calculated using two-tailed Student’s t-test. Previously, BRCA1/2 mutational status was shown to be associated with olaparib response in breast cancer and ovarian cancer patients (13,59). Analysis of tumor responses to olaparib versus BRCA1/2 mutation status showed that BRCA1/2 mutational status alone was not sufficient to predict responses to olaparib (Figure 1C), with some BRCA1/2 mutant models not responding to treatment (i.e. PDX81), whereas olaparib-sensitive models also included BRCA1/2 wt tumors (PDX37, PDX176, PDX67, PDX174) (Figure 1C). We next assessed if the level of genomic instability was associated with response to olaparib. Whereas all three PDX models with profile A did not respond to olaparib, 8/12 models with CNA profile D or E responded to olaparib (Figure 1D), underscoring that CNA profiles are not sufficient to separate olaparib-sensitive from olaparib-insensitive models (Figure 1D). For diagnostic purposes in breast cancer, a BRCA1 and BRCA2 classifier was developed to identify tumors with a ‘BRCA1-like’ or ‘BRCA2-like’ phenotype (60). Application of the ovarian adjusted BRCA1/2 classifiers to our selected ovarian PDX models, identified 10 out of 15 models as a ‘BRCA1- or BRCA2-like’ tumor (Table 1). However, 3 of these 10 ‘BRCA1/2-like’ models (i.e. PDX56, PDX143, PDX191) did not respond to olaparib, whereas 1 PDX model (PDX176) was not identified as ‘BRCA1/2-like’ but did respond to olaparib (Figure 1E). Combined, these data show that neither the presence of a genomic BRCA1/2 alteration, nor CNA levels or BRCA1/2-like classifier scores adequately predicted response to olaparib in this cohort of ovarian cancer models. Functional assessment of DNA repair in relation to olaparib response in ovarian cancer PDX models To test whether the functionality of DNA repair is associated with a response to olaparib, we analyzed two genome maintenance functions that have been attributed to BRCA1 and BRCA2, specifically the ability to recruit RAD51 to sites of DNA damage (42), and the ability to protect stalled replication forks (6,7).First, we conducted the RECAP (REpair CAPacity) assay, in which the ability of cells is assessed to recruit the downstream HR repair component RAD51 to irradiation-induced DNA damage foci (42,44). RAD51 foci formation has been previously used to identify HRD tumors and was previously shown to be associated with treatment response (61,62). Freshly isolated PDX tumor tissue from 15 models was irradiated ex vivo and the formation of RAD51 foci was subsequently visualized using immunofluorescence (Figure 2A and B and Table 1). Because HR is only employed in proliferating cells, the cell cycle-regulated protein geminin (GMMN) was included as a marker for S/G2 cells. Approximately 7 out of 15 PDX models showed RECAP positivity, as judged by at least 20% of geminin-positive cells showing RAD51 foci (Figure 2C). We observed a statistically significant relation between RECAP positivity and response to olaparib, with RECAP-negative tumors showing a decrease in tumor size upon treatment (P= 0.0002, Figure 2D). Also, when percentages of RAD51-positive tumor cells were used as a continuous variable, a strong correlation was observed between HR repair capacity and response to olaparib (r = 0.8247, P= 0.002, Figure 2E). RECAP assay predicts olaparib response in ovarian PDX models. (A) Schematic workflow of the RECAP assay. Freshly isolated PDX tumor tissue was cut into small pieces and ex vivo irradiated with 5 Gy. Tumor pieces were incubated for 3h at 37°C and subsequently fixed, stained for RAD51 and Geminin, and microscopically analyzed. Amounts of Geminin+ cells with ≥5 RAD51 foci were assessed. (B) Representative images of RAD51/Geminin staining of irradiated tumor tissue of PDX187 (RAD51+, HR proficient) and PDX37 (RAD51−, HRD). (C) Bars represent percentages of RAD51+/Geminin+ cells of individual PDX models. The dotted line marks the threshold of ≤ 20% Rad51+/Geminin+ cells, indicating HR deficiency. Boxes show different characteristics of each PDX model. Data are shown as mean ± SEM of at least three different tissue slides. (D) Olaparib response was plotted as mean tumor size at day 28 in olaparib-treated mice normalized to mean tumor size in solvent-treated mice. Each dot represents one PDX model. HR proficient (>20% Rad51+/Geminin+ cells) PDX models were separated from HRD (≤20% Rad51+/Geminin+ cells) PDX models. Data are shown as mean ± SD of different PDX models. P-value was calculated using two-tailed Student’s t-test. (E) Percentages of RAD51+/Geminin+ cells and mean percentages of tumor growth from olaparib-treated mice versus solvent-treated mice were plotted. Each dot represents one PDX model. Correlation coefficients were calculated using Pearson, and linear regression was plotted. The ability of cells to protect their replication forks is also relevant for PARPi sensitivity, as trapping of PARP onto DNA by PARPi results into replication fork stalling (63). Therefore, we analyzed the ability of tumor cells to protect stalled replication forks, which has been shown to depend on various HR proteins and was reported to be related to olaparib response (6,16). Replication kinetics and the ability of cells to protect stalled replication forks can be assessed with the fiber technique. To this end, we were able to successfully perform ex vivo DNA fiber analysis on freshly isolated tumor tissue of 13/15 PDX models. Tumor tissue was dissociated into single cells and subsequently incubated with synthetic nucleotides CIdU and ldU (Figure 3A). To deplete the nucleotide pool and stall replication forks, cells were then treated with a high dose of HU. Finally, DNA was isolated and spread onto glass slides, and the incorporated synthetic nucleotides were visualized to quantify the length of replication tracts within DNA fibers. To confirm that measured fibers were originating from tumor cells, dissociated cells used for fiber analysis were simultaneously incubated with EdU to identify proliferating cells and stained for cytokeratin to identify tumor cells. For 10 PDX models, the percentage of Edu-positive cells that were also cytokeratin-positive were analyzed, which the proliferating cells almost exclusively reflected tumor cells (Supplementary Figure S4A and B). As a readout for the ability of tumor cells to protect replication forks from HU-induced degradation, IdU/CldU ratios in HU-treated cells were normalized to the untreated setting, with a value of 1 reflecting fully protected replication forks, and a value of 0 reflecting no fork protection (Figure 3B, with all individual IdU/CldU ratios in -HU and + HU conditions in Supplementary Figure S4C). A substantial range in the ability to protect replication forks was observed between PDX models (Figure 3B). Surprisingly, PDX81 showed the lowest degree of replication fork protection, but had a heterozygous BRCA1 deletion and was classified as genomically stable (CNA profile A) (Figure 3B), whereas PDX195 and PDX193 showed defective HR as judged by loss of RAD51 recruitment, were genomically instable (CAN profile D and E) respectively, and harbored BRCA1 gene mutations, but did not show a fork protection defect. When normalized IdU/CldU ratios were used as a continuous variable, a slight negative relation between replication fork protection and response to olaparib was observed (r = −0.3872), which was not statistically significant (P = 0.1911, Figure 3C). As an additional readout for replication stress, we calculated the overall replication speed (Figure 3D). Similar to normalized IdU/CldU ratios (Figure 3B and C), analysis of overall replication speed showed a substantial degree of variation between PDX models and did not show an association with CNA profiles or BRCA1/2 mutation status (Figure 3D and E). Also, overall replication speed was not correlated with response to olaparib (r = −0.1980, P= 0.5167, Figure 3E). Analysis of replication fork stability and replication speed does not correlate with olaparib responses in ovarian PDX models. (A) Schematic workflow of ex vivo DNA fiber analysis. Freshly isolated tumor tissue was dissociated into single cells and incubated with synthetic nucleotides CldU and IdU. Cells were subsequently treated with or without HU to stall replication forks. DNA was then isolated, stretched and fixed onto glass slides. Incorporated synthetic nucleotides were visualized, and length of DNA fibers was quantified. (B) The ratio of IdU/CldU fiber lengths upon HU treatment were normalized to the untreated condition. Bars represent normalized ratios per PDX model. A ratio of <1 (dotted line) represents replication fork degradation. Boxes indicate different characteristics of each PDX model. (C) Normalized IdU/ClDU ratios and mean percentages of tumor growth from olaparib-treated mice versus solvent-treated mice were plotted. Each dot represents one PDX model. Correlation coefficients were calculated using Pearson and linear regression was plotted. (D) Replication fork speed (kb/min) of CldU fibers was calculated as mentioned in the methods. Data are shown as median of all measured fibers per PDX. Each dot represents one fiber. Boxes indicate different characteristics of each PDX model as for panel (B). (E) Median fiber speed and mean percentages of tumor growth from olaparib-treated mice versus solvent-treated mice were plotted. Each dot represents one PDX model. Correlation coefficients were calculated using Pearson and linear regression was plotted. Despite the low number of PDX models compared to clinical studies, ‘receiver operating characteristic’ (ROC) curves from genomic (Supplementary Figure S4D) and functional analyses (Supplementary Figure S4E) demonstrated that the RECAP assay could predict in vivo olaparib response with an AUC of 1. Additionally, the optimal cut-off value of the RECAP result (between 18.15 and 30.02%) determined by the ROC curve confirms our 20% cut-off used that was previously determined (42). Thus, in contrast to replication fork protection or replication speed, a functional read-out of HR repair capacity based on RAD51 recruitment was associated with response to olaparib in this cohort of ovarian cancer PDX models. Mutations in olaparib-sensitive PDX model Since a number of PDX models showed response to olaparib that could not be explained by BRCA1/2 alterations, we analyzed a panel of 226 genes that were previously demonstrated to be involved in DNA repair (Supplementary Table S1). Panel sequencing was performed on 10 PDX models (indicated in Table 1), including olaparib-sensitive PDX models without BRCA1/2 alterations. Candidate variants were filtered for germline variant frequencies found in patients and controls. Concordant with our first genomic DNA analyses, all BRCA1/2 mutations were identified in PDX81, PDX177, PDX189 and PDX195 (Table 2). Besides the identified BRCA1/2 variants, 19 additional variants were identified in olaparib-sensitive models (Table 2). Furthermore, 7/10 models showed mutated TP53 and two out of three TP53 wt samples had inactivated PTEN and mutant MSH2/3 (data not shown). PDX112 did not harbor any of these alterations and was also sequenced as a BRCA1/2 wt control model with CNA profile A. One particular variant in the olaparib-sensitive, BRCA1/2 wt PDX37 caught our attention; a single-nucleotide variant (SNV) in MUS81, altering arginine on position 496 to glutamine (R496Q, highlighted in Table 2). The underlying c.1487G > A mutation was homozygous, and resides in a highly conserved region close to the ERCC4 binding domain, within a predicted Hef domain (Figure 4A and B). Furthermore, MUS81 loss has previously been shown to cause olaparib sensitivity, including in genome-wide CRISPR screens and was shown to cause olaparib sensitivity when depleted in ovarian cancer cell lines (64,65). Using U2OS cells, we could confirm that MUS81 inactivation leads to sensitivity to olaparib and cisplatin (Supplementary Figure S5A and B), albeit without loss of RAD51 foci formation capacity. To test the effect of R496Q-MUS81 on PARPi sensitivity, we used eHAP1 MUS81 knockout cells (Figure 4C), which were transduced with wt-MUS81 or R496Q-MUS81 (Figure 4C). As expected, sHAP1 MUS81KO cells showed increased olaparib sensitivity, which was largely rescued by expression of wt-MUS81 (Figure 4D). In contrast, expression of R496Q-MUS81 only showed a partial rescue of olaparib sensitivity (Figure 4D). Of note, the expression levels of R496Q-MUS81 were slightly lower than those of wt-MUS81, possibly pointing toward R496Q-MUS81 being less stable. Analysis of MUS81 R496Q variant in the context of PARPi sensitivity. (AandB) Schematic depiction of functional domains (A) and structure of MUS81 (B) with R498 located in the Hef domain at the surface of the protein. (C) Immunoblot analysis of eHAP MUS81wt, MUS81KO and eHAP MUS81KO reconstituted with MUS81WT or MUS81R498Q. Immunoblot shows MUS81 and GAPDH protein levels. (D) eHAP cells described in C cells were treated with indicated concentration of olaparib. Percentages of surviving cells were measured by MTT. Data are presented as mean ± SEM of five replicates. (E) Quantification of AUC of cells treated as described in (D). P-values were calculated using a Student’s t-test Discussion In this study, we associated in vivo olaparib PARPi responses to genomic features related to HRD, functional HR assessment and replication fork stability in a cohort of ovarian cancer PDX models. Although RAD51 foci formation has been previously described as a tool to identify HRD tumors, and has been studied in relation to treatment response, to our knowledge, this is the first study in which both RAD51 formation and replication fork protection is analyzed in relation to in vivo PARPi response in patient-derived models of ovarian cancer. In our cohort of ovarian cancer PDX models, the presence of BRCA1/2 alterations, including BRCA1 promoter methylation, was not significantly associated with in vivo olaparib response in the tested cohort of PDX models. Also, only a subset of genomically unstable tumors, as assessed by CNA analysis, responded to olaparib. These findings underscored the need for alternative or additional selection methods for PARPi eligibility. Functional testing of replication fork stability or replication speed using DNA fiber analysis on freshly isolated tumor tissue did not correlate to in vivo olaparib responses. In contrast, the RECAP assay, which tests HR functionality as determined by the formation of RAD51 foci in ex vivo tumor tissue, fully predicted in vivo olaparib response. Several PDX models were identified that were defective for RAD51 foci formation and were responsive to PARPi, while they did not harbor BRCA1/2 alterations. These results further support the use of the RECAP method to identify patients that may benefit from PARPi, beyond those with BRCA1/2 mutant tumors. Clinical trials previously demonstrated that PARPi can be beneficial in patients with breast or ovarian cancers lacking BRCA1/2 mutations (33,41), and other cancer types (66). So, there is a clear need for a robust biomarker of HR functionality that correlates with PARPi response. To this end, various HRD tests have been developed, focused mostly on genomic features. The myChoice HRD test combines BRCA1/2 mutation status with different measurements of genomic instability, but was unable to predict responses to the PARPi niraparib in ovarian cancer (33). HRDetect is based on algorithms on whole-genome sequencing profiles from BRCA1/2-mutated breast cancers and was able to detect HRD tumors that responded to platinum-based chemotherapy (34). Reported studies in which HRDetect is used to predict responses to PARPis are still lacking. Despite the fact that these genomic analyses are relatively easily applicable in the clinic, they do not reflect HR functionality at time of treatment decision making. This is relevant, because functionality of the HR pathway can be restored in BRCA1/2 mutant tumors through secondary mutations which cause PARPi resistance (30,37). Such mutations can arise due to previous treatments with DNA damaging agents such as chemotherapy or during PARPi treatment (67). Our findings are in line with other reports, in which functional testing of HR by RAD51 foci formation is able to detect HR deficiency in breast cancer (44). The RECAP assay was used in an extensive cohort of primary breast cancer tissues (n = 148) and identified 19% of these samples to be HRD, of which seven samples were non-BRCA1/2-related (44). In this latter study, however, correlation with clinical treatment response was lacking. The relevance of assessing HR functionality by detection of RAD51 foci was underscored in BRCA1/2-mutated breast cancer PDX models, in which HR function was restored and therefore caused PARPi resistance (46). Importantly, the RECAP assay was able to identify HR restoration, despite various underlying mechanisms. Also, the RECAP assay was able to detect restoration of HR in BRCA1 mutant metastatic breast cancer upon treatment with platinum-based and PARPi treatment (47). In the RECAP assay, the formation of RAD51 foci is assessed by immunofluorescence upon ex vivo irradiation. Although this creates a good dynamic window to detect HR defects, it comes with logistic challenges, including the processing of fresh tumor tissue and the availability of a radiation source. Interestingly, recent studies analyzed endogenous RAD51 foci in untreated paraffin-embedded samples. In this approach, RAD51 assessment was successful in identifying olaparib-sensitive breast cancer PDX models, beyond BRCA1/2 mutated tumors (45). Whether this approach is also feasible for ovarian cancer tissues remains unclear. Of note, PARPi resistance might also occur independently of HR restoration, involving restored replication fork protection (16) or suppression of ssDNA gaps (19). However, using non-transformed human epithelial cell lines, primarily the role of BRCA2 in HR was responsible for maintaining cell viability and prevention of replication stress, and not its role in replication fork protection (68). Moreover, the role of BRCA2 in HR, but not fork stabilization or ssDNA gap suppression was recently shown to determine PARPi sensitivity in genetically engineered mouse models with separation-of-function Brca2 mutants (29). In line with these findings in engineered systems, in our PDX cohort the HR capability of tumor cells but not the ability to protect replication forks was associated with in vivo olaparib response. Interestingly, two PDX models (PDX81, PDX189), with a BRCA1 and BRCA2 mutation respectively, were classified as genomically stable, as based on CNA profile (profile A and B) and did not show BRCAness based on BRCA1/2-like scores (score 0.00 for BRCA1-like and 0.07 and 0.09 for BRCA2-like, respectively). Furthermore, PDX81 did not respond to olaparib in vivo. The observation that these PDX models did not behave as ‘BRCAness’ tumors, might partially be explained by these tumors being derived from primary endometrioid ovarian tumors. Furthermore, the identified BRCA1/2 mutations had an allele frequency of approximately 50%, suggesting that these are heterozygous mutations without LOH. Finally, both tumors were classified as MSI-high, caused by MLH1 promoter methylation in one model (PDX189). This may have resulted in the presence of a heterozygous BRCA1/2 mutation as a passenger mutation, without a BRCAness phenotype. Unfortunately, PDX189 was not available for in vivo olaparib response analysis. For PDX181, no underlying cause of MSI was identified. We finally aimed to find the underlying cause of HRD in PDX models sensitive to olaparib without harboring a BRCA1/2 alteration (n = 4). We identified several interesting homozygous mutations in olaparib-sensitive PDX models, as a possible mechanism underlying HRD. We identified a variant in MUS81, which was not able to rescue PARPi sensitivity when compared to wt MUS81. However, loss of MUS81 did not interfere with RAD51 recruitment, making it unclear if the MUS81 variant in PDX37 is responsible for olaparib sensitivity, while also indicating that additional defects in PDX37 are responsible for the RAD51 defect. Moreover, the slightly lower expression level of R496Q-MUS81 may indicate this this variant is less stable at the protein level. Future research into MUS81 variants, and analysis of other genomic aberrancies is warranted to reveal which mutations underlie HR deficiency and olaparib sensitivity in these models. In summary, we showed that functional testing of HR with the RECAP assay correlates to in vivo olaparib response in a cohort of ovarian cancer PDX models. Surprisingly, the presence of BRCA1/2 alterations, CNA profile or ex vivo assessed replication fork protection did not correlate to in vivo olaparib response in ovarian cancer PDX models. Several PDX models were identified as HRD and PARPi sensitive without carrying a BRCA1/2 alteration. The RECAP assay warrants further investigation in clinical trials to assess its predictive potential for in vivo response to PARPi therapy to increase the population of patients that might benefit from PARPi therapy. Limitation of the study This study has limitations. First, we only correlated in vivo PARPi sensitivity of PDX models to HR function and replication fork stability analysis. We did not analyze ssDNA gaps, a recently identified mechanisms of PARPi sensitivity in BRCA1/2 mutant cancer cells. Therefore, in the current study we cannot draw conclusions on how ssDNA gap determination would perform in comparison to HR functionality in predicting PARPi response. Second, analysis of replication fork velocity in PDX material showed significant variation within tumor samples, which we previously also observed in DNA fiber analysis of primary breast cancer tissues (69). Using patient-derived tumor tissues, iso-genic controls are lacking to determine a threshold for defective replication fork protection, which may negatively impact the interpretability of these assays. Thirdly, we identified a MUS81 variant in PDX37, which conferred partial PARPi sensitivity when introduced in a cell line model. However, loss of MUS81 does not lead to a defect in RAD51 loading, indicating that the MUS81 variant is not responsible for all phenotypes of PDX37. Supplementary Material Acknowledgements Author contributions: M.A.T.M.v.V. and F.T. conceived and initiated the project. F.T. and V.O.N.T. performed in vivo and ex vivo experiments. E.H.R., P.M.N. and R.D.D. performed genomic BRCA1/2 and CNV analyses. M.J., P.Z. and Z.K. performed targeted DNA sequencing. S.d.J. and G.B.A.W. established and provided PDX models. E.W.D. analyzed H&E stainings from PDX tumors. F.T., V.O.N.T., E.H.R., P.M.N., M.J., P.Z., Z.K. and M.A.T.M.v.V. analyzed data. F.T. and M.A.T.M.v.V. wrote the manuscript. All authors provided feedback on the manuscript before submission. We thank Frans Hogervorst and Rob Plug for technical support; Stephen West and Kasper Fugger for sharing reagents; members of the Medical Oncology Department for discussions and feedback on the manuscript; and Marta Kalousova and Marketa Safarikova for providing technological capacity for NGS analyses. Contributor Information Francien Talens, Department of Medical Oncology, University Medical Center Groningen, University of Groningen, Hanzeplein 1, 9713GZ Groningen, the Netherlands. Vivian Oliviera Nunes Teixeira, Department of Medical Oncology, University Medical Center Groningen, University of Groningen, Hanzeplein 1, 9713GZ Groningen, the Netherlands. Yannick P Kok, Department of Medical Oncology, University Medical Center Groningen, University of Groningen, Hanzeplein 1, 9713GZ Groningen, the Netherlands. Mengting Chen, Department of Medical Oncology, University Medical Center Groningen, University of Groningen, Hanzeplein 1, 9713GZ Groningen, the Netherlands. Efraim H Rosenberg, Department of Pathology, Netherlands Cancer Institute, Plesmanlaan 121, 1066CX Amsterdam, the Netherlands. Rashmie Debipersad, Department of Pathology, Netherlands Cancer Institute, Plesmanlaan 121, 1066CX Amsterdam, the Netherlands. Evelien W Duiker, Department of Pathology and Medical Biology, University Medical Center Groningen, University of Groningen, Hanzeplein 1, 9713GZ Groningen, the Netherlands. Nathalie van den Tempel, Department of Medical Oncology, University Medical Center Groningen, University of Groningen, Hanzeplein 1, 9713GZ Groningen, the Netherlands. Marketa Janatova, Institute of Medical Biochemistry and Laboratory Diagnostics, First Faculty of Medicine, Charles University and General University Hospital in Prague, Katerinska 32, 128 00 Prague, Czech Republic. Petra Zemankova, Institute of Medical Biochemistry and Laboratory Diagnostics, First Faculty of Medicine, Charles University and General University Hospital in Prague, Katerinska 32, 128 00 Prague, Czech Republic; Department of Pathophysiology, First Faculty of Medicine, Charles University, U Nemocnice 5, 128 00 Prague, Czech Republic. Petra M Nederlof, Department of Pathology, Netherlands Cancer Institute, Plesmanlaan 121, 1066CX Amsterdam, the Netherlands. G Bea A Wisman, Department of Gynecology and Obstetrics, University Medical Center Groningen, University of Groningen, Hanzeplein 1, 9713GZ Groningen, the Netherlands. Zdenek Kleibl, Institute of Medical Biochemistry and Laboratory Diagnostics, First Faculty of Medicine, Charles University and General University Hospital in Prague, Katerinska 32, 128 00 Prague, Czech Republic; Department of Pathophysiology, First Faculty of Medicine, Charles University, U Nemocnice 5, 128 00 Prague, Czech Republic. Steven de Jong, Department of Medical Oncology, University Medical Center Groningen, University of Groningen, Hanzeplein 1, 9713GZ Groningen, the Netherlands. Marcel A T M van Vugt, Department of Medical Oncology, University Medical Center Groningen, University of Groningen, Hanzeplein 1, 9713GZ Groningen, the Netherlands. Data availability The data underlying this article will be shared on reasonable request to the corresponding author. Supplementary data Supplementary Data are available at NAR Cancer Online. Funding The Dutch Cancer Society [2014–7048 to M.A.T.M.v.V.], MZ CR NU20-03-00016, RVO-VFN64165, MSMT CR LX22NPO5102] to Charles University. Conflict of interest statement. M.A.T.M.v.V. has acted on the Scientific Advisory Boards of Nodus Oncology and RepareTx, unrelated to this work. All other authors declare no competing. References Associated Data Supplementary Materials Data Availability Statement The data underlying this article will be shared on reasonable request to the corresponding author.